Re:Udp/tcp connection using pure bash
by viraptor on Sunday February 19, 2012 @09:35AM
(#3609419)
While it's an interesting trick - With no proper timeout / error control, I'd rather see the already bloated bash to be a couple of bytes shorter, than to include that feature. There's 'nc' when it's needed. We're getting quite far from the unix's 'one tool - one simple task' idea.
Re:Udp/tcp connection using pure bash
by moe on Sunday February 19, 2012 @11:42AM
(#3609664)
Agreed. These toy-features are especially bitter when you look at how little bash the language has evolved in the past decade.
Pretty much every system out there drags a significant baggage of shell-scripts along. Yet there's been barely any improvement on the arcanest of all syntaxes. We keep on painstakingly reinventing ever the same control structures (half-broken lockfile mechanisms, retry loops, miserable support for pid/pgroups, awkward subprocess/fd handling etc. etc.) because for some reason nobody bothers to finally bake them into the shell.
It's become so bad that most teams seem to resort to using a different scripting language for complex tasks nowadays. But that comes with its own hairball of problems and imho the de-facto shell-scripting language simply should not fall behind its own problem domain so far that the preferred approach becomes to not use it.
Re:Udp/tcp connection using pure bash
by veyron on Sunday February 19, 2012 @12:02PM
(#3609699)
Lets imagine for a moment that cross-platform backwards compatibility is a high priority, and that every change leads to a frustrated group of users running scripts from many years ago. How could you make a destructive change?
As an example, this works in bash 3.2 but not 4.0:
Re:Udp/tcp connection using pure bash
by moe on Sunday February 19, 2012 @12:14PM
(#3609731)
How could you make a destructive change?
Well, the same way that everyone else does it: by versioning.
The simplest approach would be to have /bin/bash2, /bin/bash3 etc., with /bin/bash defaulting to the last (current) unversioned bash release.
Then when you write a script that depends on a new feature you'd reference #!/bin/bash3 or whatever. More complex schemes are possible but probably not needed here. It's not rocket science.
And as your example illustrates: This kind of versioning is actually needed even in the current (almost stale) development-mode of bash. Its absence is the reason why many people still target sh as the lowest common denominator...
Re:Udp/tcp connection using pure bash
by nknight on Sunday February 19, 2012 @07:07AM
(#3609180)
Er, what OS is this referring to? My Linux boxes don't have a /dev/tcp or /dev/udp. This sounds like a Plan9ism, but IIRC Plan9's network interface shows up under /net, not /dev.
EDIT: Ah, this is truly a Bash feature, some sort of internal hackery. It doesn't depend on the actual existence of tcp or udp devices.
Pretty much every system out there drags a significant baggage of shell-scripts along. Yet there's been barely any improvement on the arcanest of all syntaxes. We keep on painstakingly reinventing ever the same control structures (half-broken lockfile mechanisms, retry loops, miserable support for pid/pgroups, awkward subprocess/fd handling etc. etc.) because for some reason nobody bothers to finally bake them into the shell.
It's become so bad that most teams seem to resort to using a different scripting language for complex tasks nowadays. But that comes with its own hairball of problems and imho the de-facto shell-scripting language simply should not fall behind its own problem domain so far that the preferred approach becomes to not use it.
As an example, this works in bash 3.2 but not 4.0:
in 3 it expands to whereas in 4 it expands toWell, the same way that everyone else does it: by versioning.
The simplest approach would be to have /bin/bash2, /bin/bash3 etc., with /bin/bash defaulting to the last (current) unversioned bash release.
Then when you write a script that depends on a new feature you'd reference #!/bin/bash3 or whatever. More complex schemes are possible but probably not needed here. It's not rocket science.
And as your example illustrates: This kind of versioning is actually needed even in the current (almost stale) development-mode of bash. Its absence is the reason why many people still target sh as the lowest common denominator...
read line <&456
To see what you got:
echo $line
EDIT: Ah, this is truly a Bash feature, some sort of internal hackery. It doesn't depend on the actual existence of tcp or udp devices.